Computer Engineering — Good to Know: Rules of Thumb & Mastery Guide

Computer Engineering — Good to Know: Rules of Thumb & Mastery Guide

This guide covers everything in the computer discipline catalogue without a standalone study file: the standard-practice and niche credentials, the standards and regulations you must know cold, the order-of-magnitude numbers for networks and data centers, and the veteran knowledge no exam tests.

1. The standard-practice & niche credentials

Entry and foundation

• CompTIA A+ — hardware, OS, troubleshooting basics. Who needs it: help-desk / field technicians starting out. When: first 6 months of an IT career. Honest take: near-zero value past entry level, but it's the on-ramp that proves you can diagnose systematically.
• CompTIA Network+ — vendor-neutral networking fundamentals (addressing, protocols, troubleshooting methodology). Who needs it: people approaching CCNA who want a gentler first pass, or non-Cisco shops. When: before CCNA, or instead of it if you'll never touch Cisco CLI. It teaches concepts; CCNA teaches the craft.
• CompTIA Security+ (SY0-701) — entry security baseline, also listed under software. For computer engineers: the prerequisite vocabulary for CISM/CISSP and for any security-adjacent infrastructure role.

The Juniper track (the non-Cisco path)

• JNCIA-Junos — Juniper's associate: Junos OS basics, routing fundamentals on Juniper gear. Who needs it: engineers in Juniper shops (service providers, large enterprises, cloud builders). When: as the alternative entry to CCNA when your employer is Juniper-based.
• JNCIP-ENT — professional-level enterprise routing/switching on Junos (OSPF, BGP, multicast, EVPN/VXLAN). Who needs it: senior engineers in Juniper environments. When: after JNCIA + real Junos experience. The meta-point: Junos's commit/rollback configuration model (commit confirmed, candidate configs) is operationally superior to IOS's immediate-apply — knowing both makes you a better engineer regardless of vendor.

Cloud networking specialties

• AWS Certified Advanced Networking – Specialty — deep AWS networking: Direct Connect, Transit Gateway architectures, hybrid DNS, network security at scale, multi-account network design. Who needs it: network engineers whose infrastructure is now AWS. When: after SAP-C02 or CCNP-level networking + AWS experience. This is where the CCNP guide's theory meets the SAP-C02 guide's services.
• Microsoft Certified: Azure Network Engineer Associate (AZ-700) — Azure networking: VNets, ExpressRoute, Azure Firewall, Private Link, traffic management. Who needs it: Azure-shop network engineers. When: after AZ-104.

Security management (the governance layer)

• CISSP — canonical shared guide: ../shared/cissp-study-guide.md. For computer engineers: take it when moving from hands-on network/security engineering toward architecture or management. It's governance, not packet craft.
• CISM (Certified Information Security Manager) — ISACA's management cert: governance, risk, incident management, program development. Who needs it: security managers, GRC-adjacent engineers. When: after CISSP or 3–5 years of security operations. More management-focused than CISSP; less technical.

Wireless specialty

• CWNA (Certified Wireless Network Administrator) — vendor-neutral Wi-Fi deep dive: RF fundamentals, 802.11 mechanics, site surveys, security. Who needs it: wireless engineers (the RF side the CCNA only sketches). When: before CWNP's professional tracks, or as the wireless complement to CCNP. Veteran note: Wi-Fi is applied physics — Fresnel zones, EIRP math, and interference hunting. CWNA is one of the few certs that teaches actual physics, and it shows in the quality of its holders.

The meta-lesson

Computer engineering careers run on vendor depth + standards literacy. The Cisco track (CCNA→CCNP→CCIE) is the deepest single-vendor ladder in the industry; Juniper is the credible alternative; cloud networking specialties are where the jobs are migrating. RCDD owns the physical layer; Uptime Tier thinking owns the facility. No license to practice exists — but product regulation (FCC equipment authorization, export controls) and carrier obligations (CALEA, EECC) constrain what you can build and sell.

2. Standards & regulations mastery

IEEE 802.3 (Ethernet) — know cold

The wired-LAN standard family: frame format (preamble, dest/src MAC, EtherType, payload, FCS), speed variants (802.3ab 1000BASE-T, 802.3ae 10GBASE fiber, 802.3bz 2.5/5GBASE-T, 802.3bs 200/400GbE...). Know cold: auto-negotiation behavior and failure modes (duplex mismatch — the classic: one side auto, one side forced full → collisions/errors on the half side), PoE variants (802.3af 15.4W, 802.3at 30W, 802.3bt 60/90W — power budgeting per switch matters), and that Ethernet won by being good enough and cheap, not by being technically optimal.

IEEE 802.11 (Wi-Fi) — know cold

Generations: a/b/g (legacy), n (MIMO, 40 MHz), ac (5 GHz, MU-MIMO downlink), ax (Wi-Fi 6: OFDMA, BSS coloring, efficiency), be (Wi-Fi 7: 320 MHz, MLO). Know cold: 2.4 GHz has only 3 non-overlapping 20 MHz channels (1/6/11) — the #1 real-world Wi-Fi problem; 5/6 GHz is where capacity lives; co-channel interference vs adjacent-channel (co-channel is actually preferable — devices defer politely); WPA3-SAE for auth. Demonstration: site survey (predictive + active), spectrum analysis. Common violations: max-power-everything (more power = more interference, not more coverage — turn down APs and add more of them), ignoring DFS channels.

IEEE 802.1X — know cold

Port-based NAC: supplicant → authenticator (switch/AP) → authentication server (RADIUS). EAP methods: EAP-TLS (certs, strongest), PEAP (tunnel + inner auth, most common), EAP-TTLS. Know cold: 802.1X authenticates the port, then typically assigns VLAN/policy via RADIUS attributes (dynamic VLAN assignment, dACLs, SGT). MAB (MAC auth bypass) for non-supplicant devices (printers, cameras) — weaker, segment them. Common violations: deploying 802.1X without a remediation/onboarding plan (help-desk meltdown), monitor-mode forever (never enforcing).

IETF RFCs (TCP/IP suite) — know cold

The canon: 791 (IPv4), 8200 (IPv6), 826 (ARP), 792 (ICMP), 768 (UDP), 793 (TCP), 1034/1035 (DNS), 2131 (DHCP), 4271 (BGP), 2328 (OSPF). Know cold as behaviors: TCP congestion control (slow start, AIMD — the reason a single stream underperforms on long fat pipes), NAT semantics (1918 + 3021 + 2663), ICMP's role in PMTUD (and why blocking all ICMP breaks things). You don't memorize RFC numbers for the job; you internalize the behaviors they specify.

ISO/IEC 27001 — know (networking implications)

Annex A network controls: A.8.20 (network security — segmentation, segregation), A.8.21 (security of network services), A.8.22 (segregation in networks). For network engineers: this is the management-system justification for VLANs, firewalls, NAC, and logging — know how to map your technical controls to the auditor's control list.

ANSI/TIA-942 & ANSI/TIA-568 — know

Covered in depth in the RCDD and Uptime guides. The one-line mastery: 942 = what a data center facility must provide (Rated 1–4); 568 = how the cabling inside is built. Computer engineers who only know packets and not pathways design networks that can't be built.

PCI DSS v4.0 — know cold (networking implications)

Requirement 1 (firewall configurations and network segmentation — documented rulesets, reviewed 6-monthly), Requirement 11 (vulnerability scanning, pen testing). Know cold: the CDE must be segmented from everything else; "flat network with a firewall at the edge" fails; wireless in scope must be inventoried and pen-tested. Common violations: firewall ruleset sprawl (any/any rules added "temporarily" in 2019), undocumented business justifications, segmentation that exists on paper but not in the routing table (test it with actual packets).

FCC Part 15 — know cold

Radio frequency devices: Subpart B (unintentional radiators — your switch/router must not emit harmful interference; Supplier's Declaration of Conformity), Subpart C (intentional radiators — Wi-Fi APs: power limits, DFS/TPC in 5 GHz, modular approvals). Know cold: deploying wireless = operating under Part 15 rules (no harmful interference, must accept interference received); equipment authorization (FCC ID) is required before marketing in the US. For engineers: check the regulatory domain on every AP deployment (wrong domain = illegal power/channels).

CALEA — know

Communications Assistance for Law Enforcement Act: telecom carriers (and now some broadband/VoIP providers) must engineer networks to support lawful intercept. Know: it's a carrier obligation, not an enterprise one; it shapes carrier equipment requirements (mediation devices, handoff interfaces). Know it exists; you'll never configure it directly unless you work at a carrier.

European Electronic Communications Code (EECC) — know

EU framework for telecom regulation: spectrum, consumer rights, emergency communications, and (increasingly) security requirements for networks. Know: if you operate networks in the EU, the national regulator transposing the EECC sets the rules on numbering, emergency calling (112), and incident reporting. Context-level knowledge.

EU Energy Efficiency Directive (data centre rules) — know cold (if you build in the EU)

Requires data center operators to report energy performance (PUE and friends) to a European database, with efficiency expectations tightening. Know cold: PUE = total facility energy / IT energy (1.0 = perfect, ~1.5 = decent, >2.0 = legacy); the directive makes PUE a reported regulatory metric, not just an engineering brag. Design implication: efficiency (free cooling, hot-aisle containment, high-voltage distribution) is now compliance-relevant in the EU.

EAR Advanced Computing / HPC Export Controls — know

US export controls on high-performance computing: GPUs/accelerators above certain performance thresholds (total processing performance / interconnect bandwidth) need licenses for export to certain destinations (notably China/Macau). Know: this directly constrains data-center build-outs and AI cluster exports — the "which GPUs can ship where" question. Thresholds change; verify current ECCNs (3A090/4A090 family) before planning cross-border AI infrastructure. Call export counsel — getting this wrong is a criminal matter, not a design review.

3. Rules of thumb

Latency (networking reflexes):

• Speed of light in fiber: ~200 km/ms (~5 µs/km). NYC→London (~5,500 km) ≥ 28 ms one-way physics — no protocol tunes below this.
• Cross-AZ: 0.5–2 ms; cross-region US: 20–60 ms; transpacific: 100–150 ms.
• Serialization: 1500 bytes on 1 Gbps ≈ 12 µs; on 100 Mbps ≈ 120 µs; on T1 ≈ 8 ms. Below ~100 Mbps, serialization dominates voice delay budgets.
• DNS resolution: 20–100 ms typical — the hidden tax on every new connection (connection coalescing/keepalives exist for this reason).

Bandwidth & throughput:

• TCP single-stream throughput ≤ window / RTT. Default 64 KB window on 100 ms RTT ≈ 5 Mbps — window scaling (RFC 7323) is why modern transfers go faster. Parallel streams (4–8) for bulk transfer over long fat pipes.
• Rule: provision WAN at 2× measured peak; provision LAN uplinks at 4:1 oversubscription max for user access (data center: 3:1).
• iPerf is truth; speedtest.net is entertainment. Always test with the actual traffic profile.

Availability:

• 99.9% = 8.76 h/yr; 99.99% = 52.6 min/yr; 99.995% (Tier IV) = 26 min/yr. Memorize.
• Dual everything with a single management plane = dual everything with a single point of failure. The control plane is part of the availability math.
• Convergence: BFD 150 ms beats any IGP timer tuning; BGP PIC beats reconvergence.

Power & cooling (data center reflexes):

• PUE = total / IT. Every 0.1 of PUE on a 1 MW IT load ≈ 100 kW ≈ $100K+/yr (at $0.12/kWh, 8760 h). Efficiency is money.
• Rule of thumb: ~1 kW of cooling per 1 kW of IT (plus distribution losses) — the reason power and cooling are designed together.
• UPS batteries: 5–15 min at full load. Generators: 10–60 s to accept load. The gap between them is the entire UPS industry.
• Rack density: 5–10 kW/rack is normal; >15 kW needs containment/rear-door cooling; >30 kW needs liquid. Design the cooling for the actual load, not the nameplate.

Cabling:

• 90 m permanent link / 100 m channel. Copper 100 m. OM4 10G: 400 m. Single-mode: kilometers (the answer to every "how far" question beyond 400 m).
• Bend radius violations are the #1 cause of "certified but flaky" fiber. The OTDR doesn't lie; the installer does.

Estimation reflexes:

• "How many APs?" — one per 1,500–3,000 sq ft for capacity (not coverage) in offices; survey beats math.
• "How much bandwidth?" — measure peak, ×2 for growth, ×1.5 for burst headroom. Then monitor and revisit yearly.
• Subnet sizing: always leave 2× headroom (a /24 for 100 hosts today = a renumbering project in 2 years; use /23).

4. Veteran knowledge

What exams don't test:

• The 2 a.m. outage. Real troubleshooting is Bayesian: update probabilities across layers with each test, and know when to stop diagnosing and start mitigating (fail over first, root-cause second).
• Vendor TAC diplomacy. Half of expert-level networking is getting the right TAC engineer and giving them the right data (topology, configs, show tech before they ask).
• Change management. The technical change is 20%; the MOP, rollback plan, stakeholder comms, and maintenance window are 80%. Outages are caused by changes; changes are managed by process.
• Reading packet captures. Wireshark fluency — following a TCP stream, spotting retransmits, reading TLS handshakes — is the skill that separates seniors from juniors more than any cert.

Costly beginner mistakes:

1. Flat networks. "We'll segment later" = never. Segment on day one (it's 10× harder with production traffic).
2. VLAN 1 everywhere. Native/management on VLAN 1 is the default every attacker expects. Change it.
3. No out-of-band management. When the network is down and your management rides the network, you're driving to the data center. OOB (serial console, separate management network) pays for itself the first outage.
4. Ignoring the physical layer. 30% of "network outages" are layer 1 (bad cable, dirty fiber, overheating). Check physical first, always.
5. SNMP v2c in production. Plaintext community strings in 2026. Use v3 or model-driven telemetry.
6. No config backups. RANCID/Oxidized/Archiver — 30 minutes to set up, saves careers. Version-control every config.

When to call a specialist:

• RF/wireless at scale (stadiums, warehouses): call the CWNA/CWNE-level wireless specialist — RF is physics, and physics doesn't care about your CCNA.
• Data center M&E design: call the mechanical/electrical engineers — network engineers specify the IT load; M&E engineers keep it alive. The Uptime guide is literacy, not a license to design power systems.
• BGP security incidents (hijacks, leaks): call your upstreams + consider RPKI deployment with someone who's done it — route security is unforgiving of trial and error.
• Export control questions: call counsel (noted above — criminal liability territory).
• Forensics after a breach: call the IR specialists before you touch anything — your "quick look" destroys evidence.

Original ETS reference writing, verified against public sources on 2026-09-30. Thin sourcing is flagged inline. Not legal advice — confirm current editions with the issuing body.